Foxl2functions in sex determination and histogenesis throughout mouse ovary development
- José Elias Garcia-Ortiz†1, 2,
- Emanuele Pelosi†1,
- Shakib Omari1,
- Timur Nedorezov1,
- Yulan Piao1,
- Jesse Karmazin1,
- Manuela Uda3,
- Antonio Cao3,
- Steve W Cole4,
- Antonino Forabosco5,
- David Schlessinger1Email author and
- Chris Ottolenghi1, 6Email author
© Garcia-Ortiz et al; licensee BioMed Central Ltd. 2009
Received: 28 October 2008
Accepted: 18 June 2009
Published: 18 June 2009
Partial loss of function of the transcription factor FOXL2 leads to premature ovarian failure in women. In animal models, Foxl2 is required for maintenance, and possibly induction, of female sex determination independently of other critical genes, e.g., Rspo1. Here we report expression profiling of mouse ovaries that lack Foxl2 alone or in combination with Wnt4 or Kit/c-Kit.
Following Foxl2 loss, early testis genes (including Inhbb, Dhh, and Sox9) and several novel ovarian genes were consistently dysregulated during embryonic development. In the absence of Foxl2, expression changes affecting a large fraction of pathways were opposite those observed in Wnt4-null ovaries, reinforcing the notion that these genes have complementary actions in ovary development. Loss of one copy of Foxl2 revealed strong gene dosage sensitivity, with molecular anomalies that were milder but resembled ovaries lacking both Foxl2 alleles. Furthermore, a Foxl2 transgene disrupted embryonic testis differentiation and increased the levels of key female markers.
The results, including a comprehensive principal component analysis, 1) support the proposal of dose-dependent Foxl2 function and anti-testis action throughout ovary differentiation; and 2) identify candidate genes for roles in sex determination independent of FOXL2 (e.g., the transcription factors IRX3 and ZBTB7C) and in the generation of the ovarian reserve downstream of FOXL2 (e.g., the cadherin-domain protein CLSTN2 and the sphingomyelin synthase SGMS2). The gene inventory is a first step toward the identification of the full range of pathways with partly autonomous roles in ovary development, and thus provides a framework to analyze the genetic bases of female fertility.
Menopause, one of the most clear-cut features of female mammalian aging, ensues when the ovarian follicle pool is depleted [1–3]. Hormonal regulation of ovulation from antral follicles is well studied, and many gene mutations and disruptive agents are known to affect oocyte quality and quantity at defined stages of ovary development and maturation. However, the formation and maintenance of the follicle reserve have remained less defined [4–9]. Similarly unclear is the mechanism that connects follicle formation – perhaps the most critical stage of ovary differentiation – to sex determination, the process whereby the sexually bipotential gonad forms either a testis or an ovary [5, 10–17].
Acting with the master regulator SRY, a large number of genes are currently known to drive testis differentiation in mammals (e.g., ). By contrast, although several potent regulatory genes that act in early ovary development have been identified – including those encoding the secreted proteins WNT4 and RSPO1, and the transcription factors, DAX1/NR0B1 and FOXL2 – their relation to now classical hypotheses about the genetic basis of female sex determinantion in mammals [19, 20] is still debated, and their mechanism of action is controversial ([17, 21–24], and references therein). One striking distinction among them is that the other "early ovarian" genes are mainly expressed in somatic cells of the bipotential gonad in both sexes; but Foxl2 is only expressed in females. Furthermore, mutations in Foxl2 are notably involved in three well-defined conditions that cover the entire spectrum of ovarian pathology: premature ovarian failure with and without ovarian dysgenesis in humans ; blockage of follicle formation with secondary partial ovary-to-testis sex reversal in mice [14, 26]; and embryonic sex reversal, sometimes leading to complete XX maleness in goats .
We have further shown that in mice the combined loss of Foxl2 and Wnt4 leads to extensive gonadal XX sex reversal involving all cell lineages . A similar degree of sex reversal was recently obtained in a mouse knockout model for Rspo1 . In humans, RSPO1 is the only ovarian gene known to be associated with complete XX sex reversal , and in mice, it is at least partly required for Wnt4 expression . In fact, Foxl2 and Rspo1 are expressed independently of one another throughout fetal life [17, 21–23]. Thus, Foxl2 and Rspo1 regulate distinct female sex determining pathways and redundantly antagonize the action of testis determinants such as the transcription factor SOX9.
Like its regulatory gene Rspo1 , Wnt4 activation in the embryonic ovary is also independent of Foxl2 throughout mouse fetal life ( and this study). This indicates that studies of Wnt4 and Foxl2 can provide important complementary information about female sex determination and early ovary differentiation. We have compared ovarian transcriptomes of wild-type mice with mice lacking Wnt4, Foxl2, or both in the context of comparative analyses of all available microarray datasets for gonadal development and maturation. The results point toward highly specific candidate genes involved in gonadal sexual dimorphism, ovary aging, or both. The putative gene targets were in general consistent with further ovarian transcriptome analyses of ovaries deficient in oocytes (Kit/c-kit Wv/Wv), and were validated by corresponding changes in their expression in mouse embryos overexpressing transgenic Foxl2.
1. Overview of gonadal development and maturation by principal component analysis: convergence of the maturing ovary to a testis-like profile
To focus on ovary maturation and aging in relation to gonadal development, we compared comprehensive gene expression profiles among a set of 141 microarray gonadal samples, of which 43 come from our study and 98 were reanalyzed from public databases deposited by 4 other laboratories in addition to ours (see Methods and below, section 2). The samples spanned the life history of the mouse ovary and testis from the bipotential gonad to adulthood, including several mouse models of premature ovarian failure. In order to identify major expression profiles associated with distinct genotypes and developmental stages, we used principal component analysis (PCA), an unsupervised clustering method that does not rely on a priori class assignment. Essentially, the method first identifies the group of genes with correlated expression profiles that account for the greatest amount of variance in the dataset (the first principal component, PC1), and then sequentially looks for additional co-expression profiles that are not correlated with the first group and account for decreasing amounts of variance (see Methods).
PC1, the first principal component, i.e., the horizontal axis of Figure 1, represented ~40% of the total variance of the developmental gonadal marker genes. In the bipotential gonad, at E10–E12, as expected, sets of genes were close by in ovary and testis. Then, along PC1, developing ovaries first diverged sharply from embryonic testes and subsequently moved back to values that were increasingly similar to newborn testes (E18 to adult ovary, as labeled at the top of Figure 1A). The trend is thus in keeping with morphological indications that some features of follicle maturation involve testis differentiation-like processes (see  for a recent review).
In the case of the second axis (PC2) as well, multiple distinct pathways may be involved, but because they appear dissociated by PCA, they are likely different and relatively autonomous from those accounting for PC1. Of interest, PC2 pointed toward many genes involved in follicle formation. For example, newborn ovaries, which are enriched in primordial follicles (e.g., [1, 2]), were remarkably clustered at the top of PC2, and several genes involved in early follicle development, i.e., Lhx8, Figla, and Sohlh1 (cf. ) strongly correlated with this axis (see additional file 2C). However, follicle formation is only part of the underpinning of PC2, because some newborn ovary samples (e.g., Kit mutants) were also high-scoring even though they are known to be largely devoid of follicles.
Overall, a representative "gonadal developmental transcriptome", trained on pre- and neonatal gonadal samples by PCA, appears to discriminate the range of wild-type gonadal variation throughout life. In addition, PCA suggests that dysgenetic ovaries activate programs that may represent the anticipation of a normal maturation/aging process and which may be a causative factor for associated pathology. However, PCA does not discriminate between possible different contributions to this aging-like process, in particular the effects of meiotic germ cell depletion vs. the activation of bona fide testis-like pathways in somatic cells. In addition, the expression profiles of some genes with important gonadal functions, including Foxl2 and Sox9 (see below), correlated only weakly with PCA coordinates. In particular, they both ranked > 100th among genes that were ordered according to their degree of correlation with PC1 using the Focus classifier (see additional file 2C; consistent results were obtained with "predictive analysis of microarray", PAM , data not shown). We infer that functional studies of ovary development and maturation may be more incisive if one includes systematic comparisons to testis samples in appropriately chosen developmental models. We thus present the molecular characterization of Foxl2-null ovaries in the context of other models of ovarian dysgenesis, with special emphasis on the comparison of wild-type ovary and testis differentiation.
2.1. Foxl2knockout ovaries
Foxl2-null ("KO") ovaries relative to wild-type and in various biological contexts.
A. Down in Foxl2KO (induced by Foxl2)
Single Foxl2KO and compound Foxl2KO
Foxl2single KOand Wnt4single KO
B. Up in Foxl2 KO (repressed by, or competitive with Foxl2)
Single Foxl2 KO
Compound Foxl2 KOs
Single Foxl2 KO and compound Foxl2 KO
Foxl2 single KO and Wnt4 single KO
We first adopted a strategy to identify highly specific (though possibly indirect) targets of Foxl2, as follows.
2.2. Candidate genes for a primary role downstream of Foxl2: profiling of compound knockout models involving Foxl2
We reasoned that primary targets of Foxl2 would tend to be differentially expressed in all ovaries lacking Foxl2 either alone (from the above analysis) or in combination with the ablation of Wnt4 or Kit. At birth, these various mouse models show very different morphology, cell composition, and endocrine function, and may thus provide a stringent framework for the identification of Foxl2-dependent genes . We compiled lists for genes that were differentially expressed in Wnt4-/-Foxl2-/- and Kit-/-Foxl2-/- double knockout ovaries relative to ovaries lacking Wnt4 or Kit function but harboring a functional Foxl2 allele (see additional file 4A–B, the 20 top genes up- and down-regulated are given in Table 1). We then intersected these lists with lists of genes associated with loss of Foxl2 alone (see above). That list comprises 190 and 78 genes that were systematically down or up-regulated in the absence of Foxl2 (see additional file 4C; the 20 top genes up and down are given in Table 1, second column from right). These lists thus include highly specific genes that may mediate Foxl2 action, as follows.
Inspection of the top genes in the lists (Table 1) indicates that a large fraction of Foxl2-dependent genes have no previously assigned role in gonadal development but have been associated with either neuronal or vascular development (as indicated by asterisk or underline, respectvely, in Table 1, column 3 from the left). Some show female-specific upregulation in supporting or steroidogenic cells of embryonic gonads, including a transcription factor, ODZ4, and several plasma transmembrane proteins, including plexin (PLXNC1), the cadherin-domain containing Calsyntenin 2 (CLSTN2), and a leucine-rich repeat protein (LRRC4) (bold font, Table 1). Because some of these genes are involved in brain cortical patterning [31, 32], they might conceivably mediate the formation of the ovarian cortico-medullary axis, which is essential for follicle dynamics and reproductive longevity in mammals, and has no histoanatomical counterpart in the testis .
Top-ranking down-regulated genes also included Grip1 (glutamate receptor interacting protein 1), a nuclear repressor required for estrogen receptor alpha activity, and the gene encoding the AKR1C14 aldo-keto reductase, which can metabolize the most potent natural androgen, dihydrotestosterone [34, 35]. Foxl2-dependence of these genes during embryonic development (Figure 2) might reflect an early role for Foxl2 in regulating sex-specific steroidogenesis, which may be particularly important in higher mammals. Furthermore, starting around 16.5 dpc, upregulation of aromatase/Cyp19a1 and the nuclear receptor Nr5a2/LRH-1, related to the master regulator of steroidogenesis, Nr5a1/Steroidogenic factor-1, was strongly reduced in Foxl2-null ovaries relative to controls.
Additional hits further down the lists included sphingomyelin synthase Sgms2 (see additional file 4C). It may be a positive regulator of primordial follicle formation and/or maintenance, given that loss of a sphingomyelinase, which performs the reverse metabolic reaction, was reported to enhance primordial follicle maintenance in mice . Another hit, encoding the cytoskeletal protein BICD1 that is related to a maternal fertility factor in Drosophila, was enriched in primordial follicles of both oocytes and somatic cells (data from [37, 38]). It may complement the action of a paralogue, BICD2, that we have previously shown to be expressed in growing oocytes .
Expression levels were sometimes altered starting from 13.5 dpc, as validated by real-time PCR (Figure 2). There the levels in E13 testis, first three bars at the left of each row, are compared to levels in the ovary collected at various developmental stages in wild-type and mutant mice; the corresponding microarray data (see additional file 5) are concordant in every case. Their relevance as candidate Foxl2 targets was further substantiated by experiments in which an added Foxl2 transgene was provided to wild-type mice (see below).
The comparison of ovaries lacking Foxl2 alone or in combination with either of two other genes therefore prioritizes candidate target genes that show sensitivity to Foxl2 loss in the independent mouse models. Candidate Foxl2 targets may also be inferred by their response to multiple ovarian regulatory pathways, as follows.
2.3. Candidate Foxl2 target genes that respond to Wnt4or other ovarian pathways
Overall, Figure 3 shows that in spite of a widespread network of positive correlations among well-known ovarian genes, they are affected to a different degree in distinct pathological conditions, with a quite clear-cut separation of the main candidate targets of somatic genes Emx2/Rspo1/Wnt4/Foxl2 from those of oocyte genes Lhx8/Nobox/Figla/Foxo3 (green/pink vs red boxes, resp., in Figure 3; different colors are used for Foxl2 and Wnt4-related genes because of their partial independence, see below). This is consistent with the proposal that somatic cells and oocytes may act on largely distinct though complementary pathways to promote ovary development. The relation of these genes to testis-like markers (blue boxes in Figure 3) was more complex (see below).
Degree of concordance between lists of differentially expressed genes in several models of ovarian dysgenesis.
Figla KO (*)
KEGG analysis on gene lists for Foxl2-null or Wnt4-null fetal ovaries.
Ubiquitin mediated proteolysis
Insulin signaling pathway
ErbB signaling pathway
MAPK signaling pathway
Wnt signaling pathway
Alanine and aspartate metabolism
One carbon pool by folate
Glyoxylate and dicarboxylate metabolism
Ribosome (protein synthesis)
Ribosome (protein synthesis)
Fructose and mannose metabolism
Calcium signaling pathway
Neuroactive ligand-receptor interaction
Leukocyte transendothelial migration
Cell adhesion molecules (CAMs)
Cytokine-cytokine receptor interaction
Natural killer cell mediated cytotoxicity
Complement and coagulation cascades
Of interest, Foxl2 loss also affected Wnt-signaling. Global alterations associated with oocyte genes were much less pronounced (none scored at the 10% FDR threshold). Consistent with a complementary action of oocytes and somatic cells, loss of oocyte genes led to a mixture of Foxl2- and Wnt4-associated features (not shown).
The results thus support our previous findings that Foxl2 and Wnt4 act independently during ovary differentiation, and further substantiate the notion that they may partly antagonize each other and/or regulate mutually competitive pathways. Nevertheless, some known gonadal genes were top-ranking candidate targets of both Foxl2 and Wnt4, including Kitl and Fst (pink-green color in Figure 3 and see additional file 1D). Novel genes of potential interest that were strongly down-regulated in ovaries lacking Foxl2 or Wnt4, and thus responsive to the sum of their actions, included Gpc4 and Akr1c14 (see additional file 7A and see Discussion). Additional hits included several transcription factors, e.g., Msx1, Grip1, Tcf4, and Foxp1 (see additional file 7A). Although they scored relatively low, these genes were up-regulated in Foxl2-transgenic mouse embryos (see below). Thus all of these genes, many of them expressed in somatic cells, now become candidates for a Foxl2-dependent role in coordinating ovarian histogenesis and/or steroidogenesis.
2.4. Anti-testis roles of Foxl2 and Wnt4, and of other possible regulatory genes
We next focused on genes that were significantly up-regulated in ovaries lacking Foxl2 alone or in combination with Kit or Wnt4. They included both ovarian and testis genes. The ovarian genes included Nr0b1/Dax1, Wnt4 and Rspo1 as well as novel candidates, such as Zbtb7c; the testis genes included Sox9, Dhh, Dmrt1, and Cyp26b1 (Table 1, Figure 2, and see additional file 3 and 6). It is notable that these and other top-scoring ovarian and testis genes have important roles in sex determination and/or early sexual dimorphism (reviewed by, e.g., ). Thus, in the absence of Foxl2, the exacerbation of Foxl2-independent ovarian pathways(s) coincided with the partial de-repression of testis-like genes. This started during embryonic development, even in single knockout ovaries that are morphologically normal until birth (see Discussion). Similarly, and consistent with much earlier onset of morphological anomalies, Wnt4-null single knockout ovaries showed de-repression of some testis genes, e.g., Dhh and Inhbb; and although Foxl2 was not up-regulated in these dysgenetic ovaries by whole-organ microarray (see additional file 5), its levels remained high compared to other ovarian genes and were increased by real-time PCR (Figure 2, and see below). In addition, FOXL2 protein expression was strongly detected in the persisting cord-like and follicle-like structures (data not shown).
The expression levels of some early testis-like genes showed opposite directions of change in the absence of distinct ovarian genes. In particular, Inhbb was up in fetal ovaries lacking either Wnt4 or Foxl2, but down in ovaries lacking the oocyte gene Lhx8 and in the most masculinized phenotype that we have studied, i.e., Wnt4-/-Foxl2-/- double knockout newborn ovaries (Figures 2 and 3). Similarly, Robo1, which is up-regulated in Wnt4-null ovaries ( and see additional file 6B), showed reduced levels in ovaries lacking Foxl2 (Figure 3 and see additional file 3A). Therefore, regulation of these testis-like genes, and their impact on the ovarian phenotype, may depend on the conditional interactions of multiple ovarian factors.
Similar to loss of Foxl2 or Wnt4, loss of Lhx8 led to up-regulation of a few genes that are normally expressed in testes (e.g., Stra6 see additional file 6F). However, this effect did not extend to genes known to be involved in testis sex determination, suggesting that Foxl2- and Wnt4-null ovaries are better suited to identify novel anti-testis genes.
The thirty top-ranking genes for an action independent of Foxl2 and Wnt4.
Inactive × specific transcripts
Runt related transcription factor 1
Potassium channel tetramerisation domain containing 14
Glycosyltransferase 8 domain containing 4
R-spondin homolog (Xenopus laevis)
LPS-responsive beige-like anchor
Iroquois related homeobox 3 (Drosophila)
Synaptonemal complex protein 3
Structural maintenance of chromosomes 1B
Fibronectin type III domain containing 5
RIKEN cDNA 2010004M13 gene
Gene model 1564, (NCBI)
Leucine zipper putative tumor suppressor 1
TAF7-like RNA polymerase II, TATA box binding protein (TBP)-associated factor
RIKEN cDNA 1700007E06 gene
Zinc finger and BTB domain containing 7C
Empty spiracles homolog 2 (Drosophila)
Glycerophosphodiester phosphodiesterase domain containing 2
NADH dehydrogenase (ubiquinone) 1 alpha subcomplex, 1
Gene model 1140, (NCBI)
Leucine rich repeat protein 1, neuronal
ATPase, Na+/K+ transporting, beta 1 polypeptide
Myosin, heavy polypeptide 8, skeletal muscle, perinatal
Doublesex and mab-3 related domain, member 1C
Regulator of G-protein signaling 2
3. Foxl2transgenic mice
All female markers normally up-regulated during early ovarian differentiation, whether sexually dimorphic or not, were expressed at consistently higher levels in 13.5 dpc Foxl2-transgenic ovaries relative to wild-type littermates (21 genes tested, Figures 4B and 4C and see additional file 11). That included Rspo1, Fst, Wnt4, Dax1/Nr0b1 and Zbtb7c. In XX embryos, up-regulation overtly extended to the meiotic markers Sycp3 and Dmc1h. The status of Foxl2 candidate targets inferred from the microarray analysis of Foxl2 knockout mice is thereby validated, and there are hints of feed-forward interactions with Foxl2-independent genes, such as Wnt4, Rspo1 and Zbtb7c (see Discussion).
Furthermore, several ovarian genes were up-regulated to various extents in 13.5 dpc transgenic XY gonads, which contained severely disorganized seminiferous tubules . These changes are consistent with the induction of a limited degree of male-to-female sex reversal in Foxl2 transgenics. Among the statistically significant genes were some expressed in meiotic germ cells (Sycp3 and Dmc1h) and others in somatic cells, i.e., Zbtb7c, Clstn2, Grip1, Lrrc4 and aromatase (Cyp19a1). In particular, induction of aromatase and Clstn2 in Foxl2 transgenic XY gonads was remarkably strong relative to testes (Figure 4B). This is in agreement with previous in vitro studies indicating that Foxl2 can transactivate the aromatase promoter (e.g., [48, 49]). Conversely, 13.5 dpc Foxl2-transgenic XY gonads showed reduced expression of the male sex determining genes Sox9 and Dhh. In addition, strong repression was seen for downstream embryonic testis-specific genes of endocrine relevance (e.g., the antimullerian hormone, Amh, and Cyp11a1/P450ssc, which were reduced 3- and 18-fold, respectively). Repression was also strikingly pronounced for the two male germ cell markers tested, Xmr and Oct4, which were reduced 26- and 4-fold, respectively (Figure 4A).
A few results were unanticipated; e.g., the Foxl2 transgene down-regulated the expression of Wnt4 and Dax1/Nr0b1 while increasing expression of Inhbb. These unanticipated effects may result from an exacerbation of the inferred partial antagonism between Foxl2 and Wnt4, and possibly relate to the complex responses shown by Inhbb attendant on the loss of various ovary genes alone or in combination (see above). In any case, the preponderance of the data provide independent support for an early anti-testis action of Foxl2 – whether by delaying development of XY gonads or by promoting male-to-female sex reversal. Most important, the Foxl2-mediated disruptive effects, including the suppression of Wnt4, were confined to XY embryos. The notion of a specific and early sexually dimorphic role for Foxl2 is thereby supported.
4. Gene dosage effects in Foxl2+/-heterozygous ovaries
Loss of a single copy of Foxl2 might represent a model for the heterozygous FOXL2 mutations that affect patients. We looked at the extent to which heterozygotes expressed molecular anomalies like those seen in Foxl2-null ovaries. We focused on two stages, 16.5 dpc and 7 dpn (days post-natum), because the corresponding Foxl2-null ovaries showed well-differentiated morphological states that were either normal or well engaged in abnormal development.
In order to identify molecular anomalies associated with heterozygous loss of function, we generated oocyte- vs somatic cell gene lists enriched in fetal vs postnatal ovaries of heterozygous mice. The lists were based on published microarray data, supplemented by a list of the probes for known testis determining genes (Methods and see additional file 12). We then evaluated differential enrichment in these lists by a sensitive algorithm that detects non-random distributions in pair-wise comparisons (Gene Set Enrichment Analysis, or GSEA, ; all these results are presented in additional file 13; see Methods).
We first tested testis-determining genes (including Dmrt1, Sox9, Dhh, Fgf9, and Sf1/Nr5a1), which, as a group, were significantly up-regulated in 16.5 dpc Foxl2+/- heterozygous ovaries relative to wild-type (p < 0.03, GSEA), with levels that were comparable to those in Foxl2-null ovaries (p = 0.20, GSEA). These findings raise the possibility that loss of a single allele of Foxl2 may be sufficient for a significant though weak derepression of testis genes in the fetal mouse ovary. Nevertheless, this effect appeared to be transient. Indeed, in 7 dpn Foxl2+/- heterozygous ovaries, the microarrays evinced no significant trend toward the expression of testis markers (p > 0.2, GSEA). This contrasted with Foxl2-null ovaries, which, consistent with our previous studies showing delayed postnatal oocyte growth and partial sex reversal (see above), showed a significant enrichment of testis determining genes at 7 dpn (p = 0.02, GSEA).
Consistent with ongoing sex reversal, 7 dpn Foxl2-null ovaries also showed a sharp extensive repression of postnatal oocyte and somatic follicle cell markers compared to both Foxl2+/- heterozygous and wild-type ovaries (p < 0.001, GSEA). Compared to Foxl2-null ovaries, Foxl2+/- heterozygous ovaries exhibited normal expression of oocyte growth genes but showed clear dosage effects for somatic follicle markers. The latter were inhibited almost as much in heterozygotic ovaries as in Foxl2-null ovaries compared to wild-type (p < 0.001, GSEA). Furthermore, as expected for a dosage-related response, the inhibition in heterozygous ovaries was less strong than in Foxl2-null ovaries (p < 0.001). Top-scoring genes were often known or likely to be involved in early steps of follicle growth, i.e., Inhba, Inhbb, Cyp11a1/P450ssc, Cyp17a1, Nr5a2 (see additional file 14), as well as novel genes with a likely role in somatic cells, e.g., Odz4 (see above). Several of these genes are known to be involved in steroidogenesis. Thus, in the absence of one Foxl2 allele, strong dosage effects in somatic follicle growth genes contrasted sharply with the normal expression of oocyte growth genes.
In addition, in 16.5 dpc and 7 dpn Foxl2+/- heterozygous ovaries, we found significant up-regulation of markers for fetal ovary development, independent of whether they were germline- or somatic cell-enriched (e.g. Figla, Sohlh1, Sycp3, vs Rspo1, Irx3, Gng13, p < 0.001 and p < 0.01, resp., GSEA). This suggests that Foxl2+/- heterozygous ovaries undergo a developmental delay starting in fetal life.
We have analyzed mouse ovary development and maturation based on comparative gene expression profiling of wild-type stages and pathological conditions. First, we used an unsupervised clustering algorithm, PCA, to show that gonadal developmental genes can discriminate stages of ovarian differentiation and maturation throughout life. However, sexual dimorphism was resolved only partly, i.e., conditional on developmental time. Thus, as a second, supervised approach, we focused on Foxl2-null ovaries.
Cellular models are currently lacking for both sex determination and follicle formation. Whole-organ studies provide an alternative route to investigate underlying developmental processes, which involve complex interactions of cell types. However, whole-organ studies are particularly sensitive to biological confounds such as overt changes in cell-type composition. To circumvent such effects, we compared conditions that were associated with highly divergent ovarian phenotypes but shared a common genotype, i.e., loss of Foxl2 or Wnt4 alone or Foxl2 in combination with Kit or Wnt4 (Section 2.2). Indeed, largely distinct cell types are affected in the null ovaries that lack Foxl2, Kit or Wnt4 (i.e., somatic supporting cells, germ cells, and stroma cells, resp.). It is thus unlikely that secondary and conditional targets of Foxl2 would be similarly affected by chance in the very different phenotypes associated with these three models of ovarian dysgenesis. Thus, the lists resulting from a series of distinct dysgenetic ovary models involving loss of Foxl2 should be enriched in primary targets of this gene. Similar considerations led us to search for common putative targets of Foxl2 and Wnt4, as their deletion induces somewhat complementary phenotypes (Section 2.3). The analysis enables us to suggest novel genetic interactions and ovarian genes that are either candidate targets or act independently of Foxl2 and Wnt4. Results were validates by Foxl2 overexpression studies and the analysis was extended to Foxl2+/- heterozygotes (Sections 2.4–4).
The data lead to four main inferences.
1) Partial convergence of somatic transcriptomes in maturing ovary and testis
One inference from the present analysis is that by unsupervised principal component analysis (PCA), meiotic activity and some testis-related pathways (which include genes such as Dhh and Cyp11a1) may be dissociated from other pathways that include follicle formation (which include Lhx8 and Figla), as they map along two different axes (PC1 vs PC2). This suggests that two major pathways, one more closely related to germ cell differentiation and the other to somatic cell histogenesis, may act autonomously during gonadal development. This is consistent with some current theories on the mechanisms of sexual dimorphism; but the range of autonomous (groups of) pathways may be greater than two, as indicated by our supervised analyses (see below)
A second inference from PCA is that a large fraction of the mouse ovarian developmental transcriptome becomes progressively similar to testis after birth. As shown in Figure 1A, PCA discriminates fetal and newborn ovaries from wild-type adult gonads, which cluster with fetal-newborn testes along PC1. This supports accumulating evidence for a link between ovary maturation and a predisposition of this organ to form testis-like tissues . We found that the shift toward testes was accelerated in several knockout or mutant ovary models compared to age-matched wild-type ovaries (Figure 1B). However, PCA did not discriminate between contributions to this pattern from the loss of meiotic activity or from the activation of testis-like pathways expressed in somatic cells. In addition, some critical gonadal genes correlated only poorly with PCA coordinates. Foxl2 and Sox9 were notable examples. PCA failed to detect their standing as the current best markers for wild-type female and male supporting cells, exemplified by their striking pattern of mutually exclusive expression in a wide range of different conditions involving partial sex-reversal [14, 17, 51]. Rather than conferring variable degrees of predisposition to sex reversal, Foxl2 and Sox9 seem to act as all-or-none switches, and their mutual antagonism may be determinative in sex determination . We further tested this proposal by supervised analyses of the Foxl2-null ovary transcriptome.
2) Specific gene candidate targets of Foxl2 and comparison of ovaries lacking Foxl2 vsother early ovarian factors
By comparing single and compound knockout ovaries lacking Foxl2, we have identified and ranked 149 candidate genes for a specific role downstream of Foxl2 (top gene list in Table 1, second column from right, and see additional file 4 for the full list). KEGG pathway analysis indicated that in Foxl2-null fetal ovaries, neuronal-like pathways, migratory activity and cell-cell interactions were reduced, whereas glucose metabolism and protein synthesis were increased (Table 3 and see additional file 8). Of note, Wnt4-null ovaries showed a largely inverse metabolic profile, and we observed a significant negative correlation between the inferred candidate target genes of Wnt4 and Foxl2 (Table 2 and Figure 3). This contrasts with results when several other ovarian genes are ablated. Those mouse models show largely positive correlations, though the strength of dependence of the shared candidate targets varies, notably between somatic genes and oocyte genes (Table 2 and Figure 3; see Results). Thus, the widely divergent molecular anomalies observed in Wnt4 and Foxl2-null fetal ovaries provide an entry point to the mechanisms underlying the partly autonomous, complementary roles of these genes in ovary development.
Consistent with KEGG results, several genes that are known to be expressed by neurons and vascular cells in other systems, e.g., Calstn2, Odz4 and Lrrc4, were among the most sensitive candidate Foxl2 targets (Table 1). In addition, some steroidogenesis-related genes (Grip1 and Akr1c14) were also affected very early by loss of Foxl2. This early effect is striking considering that in the mouse ovary, active steroidogenesis occurs postnatally. Steroidogenesis is completely repressed in postnatal Foxl2-null mouse ovaries  and steroidogenic genes are strongly affected in Foxl2 transgenic mouse embryos (discussed below). Some additional Foxl2-dependent genes, expressed in both somatic cells and oocytes, may be associated with novel ovarian functions, including Sgms2 and Bicd1 (see Results).
Comparison of the positive candidate targets shared by Foxl2 and Wnt4 pointed to additional genes that may integrate the outputs of the respective largely independent pathways (Table 1 and see additional file 8). They include the genes encoding the well-known Kit ligand, KITL, the Wnt4-effector Follistatin, the proteoglycan GPC4, and several other candidate genes for novel roles in ovary development, i.e., Msx1, Grip1, Tcf4, and Foxp1 (bold characters, Table 1; see additional file 8A–D and Figure 4). All of these genes, many of them expressed in somatic cells, are enriched in fetal wild-type ovaries compared to testis and become candidates for a role in coordinating ovarian histogenesis and/or steroidogenesis.
Novel ovarian candidate target genes that we inferred to be positively regulated by Foxl2 from the comparison of single and double knockout mice were assessed in Foxl2 transgenic embryos. All were up-regulated in XX gonads, with occasional, limited upregulation in XY gonads as well (N = 21; Figures 5B and 5C and see additional file 7). These findings confirm the ability of the microarray analysis to detect bona fide targets of Foxl2, independent of whether Foxl2 regulatory action is direct or indirect. The great majority of these genes had been missed in previous less complete in vitro microarray studies (e.g., . In particular, we found limited overlap between the differentially expressed gene lists (121 of 1192 of the genes detected in vitro were identified in vivo). The expectation that genes up-regulated by Foxl2 in vitro would likely be down-regulated in the Foxl2-null ovaries (positive candidate targets), and vice-versa (negative candidate targets), was met only to some extent (p = 0.10, chi-square). Nevertheless, the overlapping hits included several biologically validated or promising genes, e.g., Nr5a2, Zbtb7c, Mmp23 and Inhbb (see additional file 15). In short, even though in vitro data included a plethora of hits that appear unrelated to the in vivo situation, some genes that overlap do have known or putatively important roles in vivo. This suggests that in vitro models may provide useful information in the study of gonadal development, though extensive filtering by comparison to in vivo data is required.
Overall, our Foxl2 transgenic model shows a generalized induction of ovary markers in XX embryos and a severe perturbation of testis markers in XY embryos, underlining a sex-specific action of Foxl2. The strongest effects of the Foxl2 transgene in XY embryonic gonads involved male germ cell genes and sexually dimorphic steroidogenic factors [the reduction of Xmr, Oct4 and P450ssc/Cyp11a1 to baseline levels and the strong upregulation of aromatase/Cyp19a1 as well as other ovarian genes, e.g., Clstn2, Lrrc4, and Zbtb7c, Figure 4]. Nevertheless, the earliest gonadal defects involving sex reveral that are observed in female steroidogenesis-deficient mice (i.e., lacking aromatase- or both estrogen receptors) occur around the time of puberty (e.g., reviewed by ). Thus, aromatase and estrogen action can only partly account for Foxl2 action, and presumably only at later stages during ovary development and maturation. The early branches of the Foxl2 pathway remain to be characterized more fully.
3) Early embryonic action of Foxl2
A striking result from the microarray analysis is the early time at which Foxl2-null ovaries showed significant anomalies, i.e., by 13.5 dpc, the time at which germ cells commit to the oocyte fate in mice. Some ovarian genes (candidate positive targets of Foxl2) were down-regulated and some ovarian and testis-like genes were consistently up-regulated (Figure 2 and see additional file 5). Thus, Foxl2-null ovaries show subtle features of partial sex reversal starting during early embryonic development. Nevertheless, levels of testis genes were low in prenatal Foxl2-null mouse ovaries (i.e., they increased by only 7–30% of the difference between testes and baseline signal in wild-type ovaries, by real-time quantitative PCR, Figure 2). This may explain why we previously failed to detect SOX9 protein expression at these stages . In any case, the early, reproducible transcriptional dysregulation of a number of testis and ovarian genes suggests that the female sex determination program is partly impaired in embryonic ovaries that lack Foxl2. These subtle alterations may predispose to overt molecular sex reversal, which occurs postnatally in the supporting cell lineage of Foxl2-null ovaries, and is accelerated and extends to all other cell types in ovaries lacking both Wnt4 and Foxl2 (see Introduction). The data also suggest widespread antagonistic or competitive interactions between Foxl2 and Wnt4 during fetal development (see Results).
The molecular anomalies detected in Foxl2-/- single knockout ovaries occur earlier than the earliest time of onset of histological sex reversal reported in mouse models so far (late fetal life in an Rspo1-null model and in the Wnt4-/-Foxl2-/- double knock-out ovaries we have studied [17, 21] and data not shown). Instead, RSPO1 mutations and Foxl2 downregulation in humans and goats, respectively, are associated with much earlier, nearly complete forms of female-to-male sex reversal [27, 28].
To reconcile differences among species, one possibility infers that other genes, possibly expressed in oocytes [15, 53], may be involved in female sex determination, and some might well be mouse-specific. Alternatively, as we have previously suggested, several autonomous ovarian gene pathways conjointly repress testis differentiation, but their autonomy may be more pronounced in mice than in higher mammals [16, 17].
We infer several candidate anti-testis genes that are expressed independently of both Foxl2 and Wnt4 by comparing gene expression patterns associated with the partial ovary-to-testis sex reversal (as observed in the corresponding single gene knockout models) to both testis and wild-type ovary profiles. Some of these candidate anti-testis genes have been partially characterized previously – e.g., Wnt9a and Irx3 (cf. [21, 54]), whereas others are novel, e.g., Zbtb7c (Table 4). Their function should become clearer with ablation studies to determine their interactions, with particular attention to any dependence on Rspo1 . Genes with restricted gonadal expression also become candidates for possible involvement, via constitutive mutations, in human patients affected by non-syndromic forms of 46, XX sex reversal or premature ovarian failure.
4) Dosage effects in Foxl2-/+heterozygotes
A fourth inference from the analyses is that anomalies in Foxl2+/- heterozygous ovaries were milder but similar to Foxl2-null ovaries. They included reduced expression of follicle genes, notably involved in steroidogenesis, and the transient coexpression of some fetal ovary and testis genes at levels higher than normal. These features suggest that loss of one Foxl2 allele might decrease the stability of ovary sex determination, and might possibly be implicated in the etiology of premature ovarian failure in heterozygous FOXL2+/- human patients .
The connection between the mechanisms of follicle formation and female reproductive competence in mammals has long been appreciated , but an involvement of embryonic anti-testis genes has only recently been suggested. Among the known anti-testis genes, Foxl2 has a special status because it is required in supporting cells throughout ovarian development ([14, 17] and this study), and its continuous expression in postnatal primordial follicles suggests a possible role in the maintenance of female reproductive capacity . However, the comprehensive analysis of available microarrays adds to the weight of evidence that additional gene pathways that may have a more limited action in time, independently promote ovarian differentiation and repress the alternative testis fate. They include at least one pathway involving Rspo1 and Wnt4 as well as novel candidate factors ( and see above). The identification of the full range of independent anti-testis genes and their most specific targets, as inferred from appropriate series of partly overlapping combinations of mutant genotypes, now becomes a focus for further investigation.
Mouse strains and RNAs
We previously reported the generation of mice lacking Foxl2 alone or in combination with Wnt4-null or Kit Wv hypomorphic alleles, as well as mice expressing a Foxl2 transgene [17, 26]. For newborn and postnatal ovaries processed with Agilent microarrays (see below), strains were a mixed background from 129S6/SvEvTac and NIH-Swiss/BC. For 15.5 dpc Wnt4-null single knockout ovaries and the respective age-matched controls , they were a mixture of C57BL/6J and the original 129 strain. For all other ovaries processed with Affymetrix microarrays, mice were a mixture of C57BL/6J, NIH-Swiss/BC and CD1. RNA samples for each genotype and developmental stage were obtained from gonads of individual mice (biological replicates), and were processed as follows.
RNA extraction of the samples and microarray platforms
For the Affymetrix platform, total RNA was extracted with the MELT enzymatic system (Ambion) followed by purification and linear amplification with the Pico Ovation kit (Nugen). For the Agilent Platform, total RNA was extracted by mechanical homogeneization and purified on RNeasy affinity columns.
For the Affymetrix MOE430 v.2 platform, all gonadal samples were assayed on three biological replicates, except for Wnt4-null mice and their age-matched controls collected at 15.5 dpc, which were studied in duplicate (that is, two mice each). Gonads from individual mice served as biological replicates; thus controls were comparable to the small numbers of double-knockout mice that were available, limited in numbers by high prenatal mortality. A step of RNA amplification was thus required. Foxl2-null ovaries and wild-type littermates were collected at 13.5 dpc, 16.5 dpc and on the day of birth (18 samples). Ovaries from newborn mice lacking Foxl2 and Wnt4, or Foxl2 and Kit, were compared to age-matched controls lacking Wnt4 or Kit but harboring a wild-type Foxl2 allele (12 samples). Testis samples were collected at 13.5 dpc and at birth. The latter were from Wnt4-null mice, which show partly defective testis differentiation. This was done in order to facilitate the assessment of the degree of sex reversal affecting newborn ovaries lacking both Wnt4 and Foxl2 .
The microarray analysis of the Foxl2 gene dosage effects was performed on Foxl2-null, Foxl2+/- heterozygous and wild-type ovaries that were obtained in triplicate at 16.5 dpc (9 samples) or in duplicate from pools of 4–8 mice at 7 days postnatum (dpn) (4 samples). The 16.5 dpc ovary samples were hybridized on the Affymetrix MOE430 v.2 platform after RNA amplificatioon (see above), whereas the 7 dpn ovaries were assayed on the Agilent 44 k mouse developmental platform without RNA amplification.
Additional Foxl2-null and wild-type newborn ovaries were processed in triplicate and analyzed on the Agilent platform, mainly for cross-platform validation purposes (see below). All our microarray datasets can be downloaded from the public web site, http://www.ncbi.nlm.nih.gov/geo, via the accession number GSE12989.
Microarray data normalization
Raw Affymetrix data (.CEL) file processing and normalization was performed by the plier program in R (at cran.r-project.org) with an additional step of scaling to the global mean.
For data from the 44 k mouse developmental array from Agilent, after background subtraction from the Agilent analyzer raw output data, calculation of the ratio of the channels was followed by normalization of each sample values against a virtual microarray dataset by local mean regression as implemented by the loess program on R (span 0.01). The reference virtual dataset contained the average expression values for each probe across all samples.
Additional available microarray datasets (available at http://www.ncbi.nlm.nih.gov/geo or http://www.ebi.ac.uk/microarray-as/ae/)
Datasets for models of (whole-organ) gonadal dysgenesis that we have accessed from the public domain are from four distinct laboratories, as follows: Emx2-/- (GSE10216),Lhx8-/- (GSE7774), Nobox-/- (GSE7775), Foxo3-/- (GSE8249) and Wnt4-/- kidney (GSE6934) ([42, 44] and unpublished data). We processed the raw data (Affymetrix MOE430 v.2 CEL files) with plier on R. Normalized datasets were directly used for PCA or analyzed to obtain lists of genes that were differentially expressed in mutant vs control ovaries with the Focus software . We thus selected the hits corresponding to a minimum Focus score of 4 under bootstrap-optimized FDR (see below), with the exception of the Foxo3-/- samples. For those only a handful of genes were obtained with the same procedure. This was consistent with the inability of a published study to identify any differentially expressed genes at birth in these datasets . Therefore, for the Foxo3-/- samples, we raised the Focus score threshold for probe selection to 6 but used no FDR; we reason that the degree of concordance between the resulting gene lists and several other gene knockout models (see Results) validated the procedure. Furthermore, a strong correlation was obtained between the gene lists obtained from the Foxo3-/- newborn ovaries and those from 7 dpn ovaries, which could be investigated with or without optimized FDR (p = 0.001; not shown). This indicates that in spite of the lack of the FDR control, the Foxo3-/- newborn ovary gene lists were enriched in specific though possibly indirect targets of Foxo3. They were thus included in comparisons with the other prenatal and newborn ovary models (see Results). Gene lists for Wnt4-/- and Figla-/- ovaries, which were obtained with different microarray platforms, were taken from the respective reports without modification [43, 45].
The datasets involving only wild-type whole-organ gonadal samples were as follows: normal ovary and testis development from 11.5 dpc to 2 dpn (GSE5334 and GSE4818) and from 11.5 to 18.5 dpc (GSE6916) from two distinct laboratories ( and unpublished data), and one adult ovary and testis each (as well as Kitl hypomorphic adult testis) from two other laboratories (GSE8249 and GSE1986,  and unpublished data).
One dataset was for oocytes from primordial through antral follicles (GSE3351, ). We used that dataset to infer lists of oocyte genes showing decreasing vs increasing expression profiles from early through growing follicles with Focus (as above), thus inferring genes expressed in fetal vs. growing oocytes, as given additional file 12. In order to obtain somatic cell-enriched genes, we subtracted the corresponding oocyte gene lists from the lists associated with parallel expression profiles (i.e., decreasing vs increasing, resp.) in whole-organ ovaries aged 0 through 14 dpn (GSE8249, ). This produced the lists of genes enriched in fetal vs postnatal somatic cells, in additional file 12.
Additional published microarrays on purified cells were as follows: Sf1/Nr5a1:GFP from ovaries and testes aged 10.5 through 13.5 dpc (E-MEXP-454, ) or 10.5–11.5 dpc (GSE3463, ), as well as Sry:GFP-positive supporting somatic cells aged 13.5 dpc (GSE4928, ). We also reanalysed a Wnt4-/- newborn kidney microarray that was deposited but is not accompanied by publication (GSE6288).
Principal component analysis
Several approaches are available to identify markers of biological interest from microarray data. Completely unsupervised approaches, which do not incorporate information on class identity, perform less well than their supervised counterparts, which instead use such information explicitly. Supervised algorithms include recent versions of discriminant analysis (such as "prediction analysis of microarray, PAM, ). The design thus favors the identification of the most strongly class-discriminatory biomarkers. However, it is not clear how the resulting informative biomarkers, besides their role in the output classifier, might be related to one another. In addition, the identification is vulnerable to any bias resulting from a priori definition of the classes. Therefore, a mixed unsupervised-supervised approach may represent a better solution.
Principal component analysis (PCA) is an unsupervised method for clustering and classification that, contrary to other unsupervised algorithms (such as hierarchical or k-means clustering), incorporates a statistical description of the datasets. In general, the purpose of PCA is to rotate data points into a new coordinate system, such that the majority of the variance in the data is distributed along the directions of a subset of the new axes. This facilitates visualization of significant patterns or trajectories within the data. In addition, the resulting axes are uncorrelated. Thus, each axis can represent the activity of clusters of genes showing a specific pattern (i.e., positively or negatively correlated expression patterns), or linear combinations thereof, and each pattern is uncorrelated to the others, thus possibly representing an independent biological process. PCA has its own drawbacks (e.g., interpretation of axes representing complex combinations of patterns of covariation may be challenging, and axes associated with small amounts of variance may be spurious). However, it can identify the major uncorrelated trends of variation in the data and quantify their contribution to the total variance.
Mixed unsupervised-supervised approaches to PCA have been proposed that restrict the dataset to the genes that may be considered more informative for a biological process of interest, by other statistical methods [58, 59]. The resulting approach, which might be termed "calibrated PCA" (because no explicit information on predefined classes is provided), was found to perform as well as as the best fully supervised classification algorithms that are currently available . Furthermore, the approach retains the ability, which is inherent in PCA, to reveal any additional unpredicted patterns that may exist within and among the known classes, thus reducing the effect of any biases at the step of class selection.
Based on these considerations, we used a general approach defining minimal "classes" as the known biological conditions, i.e., the set of replicates that were available for each combination of genotype and developmental stage under study. We obtained the lists of genes that showed the highest level of variation across such predefined classes using the Focus software (, setting the standard cutoff at Focus score 4; see below) and carried out standard PCA on the values for such genes in the samples of interest.
By carrying out PCA on the samples taken from all experiments simultaneously, the first principal components (i.e., the PCA axes accounting for the maximum variation) distinguished the experiments more than they discriminated the biological conditions; this result was obtained on rank-normalized values – that is, in spite of the use of a procedure that could help to reduce laboratory-related bias . Therefore, we restricted PCA to the samples from each single experiment (the training dataset), leading to a PCA space onto which the samples from the other experiments (the test datasets) were mapped (i.e., classified). Mapping the test dataset involved the "pseudo-inverse matrix" method computed with the svd function in R . To further reduce the effect of experimental idiosyncrasies, we obtained the developmental marker list from one dataset and trained the PCA on a distinct dataset. In particular, we used, in turn, each of the three largest developmental series available on whole-organ gonadal samples, as well as the largest dataset on somatic cells from embyonic gonads (GSE5334 and GSE4818; GSE6916; our own dataset GSE12989; and E-MEXP-454, resp., and see above). The combination of marker list and training dataset that gave the best resolution (i.e., involving the genes that were most variable across GSE5334 and GSE4818 samples as the marker list, with our dataset as the training dataset) is discussed in the main text and is reproduced in Figures 1A and 1B; further information is given in additional file 1A and 1B. The corresponding marker list is given in additional file 2B, along with a representative shorter list that showed a comparable discriminative power (at the cutoff that retained the 300 top ranking variable markers, see additional file 2A and data not shown).
We obtained the lists of genes that were most significantly associated with PC1 and PC2 of the selected PCA (see above) by dividing each PC into three equal segments between the minimum and maximum values along the coordinates, thus defining 3 classes that contained the samples mapping within the corresponding coordinates. We then used the Focus program to rank genes by the extent to which their expression levels increased or decreased through these classes (see additional files 2B–C).
Detection and ranking of differentially expressed genes
In all cases, we applied the Focus software  to the entire microarray datasets, i.e., without any pre-filtering of the probes. Focus ranks genes according to the statistical distribution of contrasts (i.e., sums of weighted means) calculated as the linear transformation of four parameters (ratios and differences of expression intensities between conditions, as well as their consistency within each condition). The weights were chosen to represent arbitrary directions of change (such as, 1 and 0 in pair-wise comparisons, or 0, 1 and 2 in three-way comparisons). These values were then normalized according to the distribution of the replicates among the distinct conditions, as implemented in Focus. We used the bootstrap algorithm implemented in Focus to select hits that were associated with at least 1.1-fold mean differences, scored higher than the default "interest score" threshold of 4 and correspondingly to an "optimized" threshold value of the false discovery rate (FDR). The Focus interest score reflects the statistical distance of each probe from the mean of the probes' distribution in terms of the combined standard deviations of the four Focus parameters presented above. The bootstrap-based algorithm selects an FDR cut-off in order to simultaneously minimize the estimated proportion of false negative errors, as inferred from the observed t-test distribution of the expression values .
The combination of a fold-ratio with an optimized FDR is intermediate between the procedures that combine a fold-ratio with either a fixed FDR or a fixed p-value cut-off (the former is widely used but may be too stringent, the latter was recommended by the Microarray Quality Control Consortium ). In addition, as can be inferred from the gene lists produced for Foxl2-null newborn ovaries vs controls analyzed with Affymetrix versus Agilent platforms, we found that using an optimal FDR led to greater specificity (i.e., a greater fraction of consistent hits) and increased sensitivity (i.e., a larger number of total hits) than the alternative methods (not shown).
In order to detect genes that were significantly associated with the Foxl2-null or Wnt4-null ovary phenotypes, we included our wild-type and heterozygous ovary samples from all developmental stages that can be considered, phenotypically, as normal controls (see additional files 3A–B and 7A–B). This was done with the purpose of spanning the range of variation that we had observed for the normal gonadal transcriptome by PCA (see above). As a complementary approach, we also identified the genes that were shared by the three pair-wise comparisons of age-matched Foxl2-null and wild-type ovaries from 13.5 dpc through birth (see additional file 3C).
Gene Ontology and KEGG analysis
We obtained the KEGG (Kyoto Encyclopedia of Genes and Genomes) and Gene Ontology classification for mouse genes from the mmu00001.keg database at ftp.genome.jp and the "GO" libraries available at http://www.affymetrix.com and, for Agilent probes, at http://www.bioconductor.org (GOENTREZID2GO function in the GO package); the analyses were performed in R (scripts available on request). The lists of differentially expressed genes between Foxl2- or Wnt4-null knockout ovaries and wild-type controls (above) were tested for any significantly different distribution across KEGG categories, i.e., by a comparing pairwise proportions using Fisher exact tests; we set a false discovery rate (FDR) threshold of 10% (see additional file 8). Similar results were obtained with GO (not shown).
Identification of candidate genes for a role independent of Foxl2 and Wnt4
We identified the genes that showed the geatest variation according to the following profile: Wnt4- or Foxl2-null ovary > wild-type ovary > testis. For this analysis, we thus included our embryonic and testis samples in the analyses of the single knockout ovaries described above. Linear contrasts implemented in Focus helped to emphasize directions of change rather than strict pattern similarity. This three-way comparison has the advantage of reducing, in the comparison between knockout and wild-type ovaries, the relevance of both the magnitude and the sign of any differences or direction of change, as long as these differences were statistically small relative to the difference between ovary and testis. Because the mutant and wild-type ovaries were chosen to be morphologically and histologically closer to one another (e.g., before oocyte loss occurred in some models) than either was to testis, the risk of selecting idiosyncratic features of the mutants (e.g. Wnt4- or Foxl2-null ovary >> wild-type ovary ~testis) was expected to be low. Thus, genes were ranked according to how strongly they maintained or exacerbated a wild-type ovary-like activity relative to testis (see additional files 9A–B, combined -by boolean intersection- into additional file 9C; the top 30 genes from additional file 9C are given in Table 2).
Using real-time PCR (TaqMan, Applied Biosystems), we assayed the same samples that were processed for the microarrays as well as additional biological samples, employing at least three biological replicates per condition. Gonads from Foxl2 transgenic 13.5 dpc embryos were also collected and compared to their wild-type littermates, following a heat-shock treatment to activate the transgene as previously reported . Normalization was performed by scaling to expression values of the succinyl dehydrogenase A subunit, Sdha. Consistent results were obtained by using another gene as the scaling factor, Gapdh. Real-time PCR for validation of the microarray data involved the following genes (and TaqMan probes): Amh (Mm00431795_g1), Apcdd1 (Mm01257559_m1), Apoa1 (Mm00437569_m1), Aromatase (Mm00484049_m1), Asb9 (Mm00518372_m1), Bax (Mm00432050_m1), Bicd1 (Mm00802208_m1), Ccdc113 (Mm00556701_m1), Clgn (Mm00515517_m1), Cyp26b1 (Mm00558507_m1), Dax1 (Mm00431729_m1), Dhh (Mm00432820_g1), Dmrt1 (Mm00443809_m1), Dmrt3 (Mm00616649_m1), Figla1 (Mm00488823_m1), Foxl2 (Mm00843544_s1), Fst (Mm00514982_m1), Gdf9 (Mm00433565_m1), Gdnf (Mm00599849_m1), Inhbb (Mm03023992_m1), Kcnd2 (Mm00498065_m1), Mid1lip1 (Mm00471535_m1), Mmp15 (Mm00485062_m1), Mmp23 (mm00488768_m1), Nedd9 (Mm00479569_m1), Nr5a2 (Mm00446088_m1), Odz4 (Mm00446088_m1), P450ssc (Mm00490735_m1), Pcdh11x (Mm01221603_m1), Plzf (Mm01176868_m1), Ptdgs (Mm01330613_m1), Rspo1 (Mm00507077_m1), Sgms2 (Mm00512327_m1), Sox9 (Mm00448840_m1), Sycp3 (Mm00488519_m1), Wnt4(ex1–2) (Mm01194003_m1), Xmr (Mm00784689_s1), Zbtb7c (Mm00520231_m1). All results were consistent with the microarray data, with the partial exception for Mid1lip1 and Pcdh11x (Figure 2 and data not shown). The latter may correspond to alternative isoforms differentially recognized by microarray vs PCR assays.
Real-time PCR for the study of embryonic gonads from the transgenic mice (Figure 4 and additional file 11) involved the following genes (and TaqMan probes): Akrc14 (Mm00506336_m1), Amh (Mm00431795_g1), Clstn2 (Mm00502574_m1), Cyp19a1 (Mm00484049_m1), Dax1 (Mm00431729_m1), Dhh (Mm00432820_g1), Dmc1h (Mm00494485_m1), Foxp1 (Mm00474845_m1), Fst (Mm00514982_m1), Gpc4 (Mm00515035_m1), Grip1 (Mm00503844_m1), Inhbb (Mm03023992_m1), Kitl (Mm00442972_m1), Lrrc4 (Mm02527970_s1), Mllt1 (Mm00452080_m1), Msx1 (Mm00440330_m1), Oct4 (Mm00658129_gH), Odz4 (Mm00496266_m1), P450ssc (Mm00490735_m1), Plxnc1 (Mm00450687_m1), Rspo1 (Mm00507077_m1), Sgms2 (Mm00512327_m1), Sox9 (Mm00448840_m1), Sycp3 (Mm00488519_m1), Tcf4 (Mm00443198_m1), Wnt4 (Mm01194003_m1), Xmr (Mm00784689_s1), Zbtb7c (Mm00520231_m1).
We thank Dawood Dudekula for help with the microarray probe annotations.We are grateful for the high quality, consistent efforts of the NIA Comparative Medicine Section in breeding, genotyping, and providing all the analyzed mice. This research was supported entirely by the Intramural Research Program of the National Institute on Aging, NIH.
- Telfer EE, Gosden RG, Byskov AG, Spears N, Albertini D, Andersen CY, Anderson R, Braw-Tal R, Clarke H, Gougeon A, McLaughlin E, McLaren A, McNatty K, Schatten G, Silber S, Tsafriri A: On regenerating the ovary and generating controversy. Cell. 2005, 122: 821-822.View ArticlePubMedGoogle Scholar
- Mandl AM, Zuckerman S: The relation of age to numbers of oocytes. J Endocrinol. 1951, 7: 190-193.View ArticlePubMedGoogle Scholar
- McGee EA, Hsueh AJ: Initial and cyclic recruitment of ovarian follicles. Endocr Rev. 2000, 21: 200-214.PubMedGoogle Scholar
- Pepling ME: From primordial germ cell to primordial follicle: mammalian female germ cell development. Genesis. 2006, 44: 622-632.View ArticlePubMedGoogle Scholar
- Hirshfield AN: Development of follicles in the mammalian ovary. Int Rev Cytol. 1991, 124: 43-101.View ArticlePubMedGoogle Scholar
- Choi Y, Rajkovic A: Genetics of early mammalian folliculogenesis. Cell Mol Life Sci. 2006, 63: 579-590.View ArticlePubMedGoogle Scholar
- Skinner MK: Regulation of primordial follicle assembly and development. Hum Reprod Update. 2005, 11: 461-471.View ArticlePubMedGoogle Scholar
- Matzuk MM, Burns KH, Viveiros MM, Eppig JJ: Intercellular communication in the mammalian ovary: oocytes carry the conversation. Science. 2002, 296: 2178-2180.View ArticlePubMedGoogle Scholar
- Epifano O, Dean J: Genetic control of early folliculogenesis in mice. Trends Endocrinol Metab. 2002, 13: 169-173.View ArticlePubMedGoogle Scholar
- Burgoyne PS: Role of mammalian Y chromosome in sex determination. Philos Trans R Soc Lond B Biol Sci. 1988, 322: 63-72.View ArticlePubMedGoogle Scholar
- Jost A, Vigier B, Prepin J, Perchellet JP: Studies on sex differentiation in mammals. Recent Prog Horm Res. 1973, 29: 1-41.PubMedGoogle Scholar
- Byskov AG: Differentiation of mammalian embryonic gonad. Physiol Rev. 1986, 66: 71-117.PubMedGoogle Scholar
- Merchant-Larios H, Taketo T: Testicular differentiation in mammals under normal and experimental conditions. J Electron Microsc Tech. 1991, 19: 158-171.View ArticlePubMedGoogle Scholar
- Ottolenghi C, Omari S, Garcia-Ortiz JE, Uda M, Crisponi L, Forabosco A, Pilia G, Schlessinger D: Foxl2 is required for commitment to ovary differentiation. Hum Mol Genet. 2005, 14: 2053-2062.View ArticlePubMedGoogle Scholar
- McLaren A: Development of the mammalian gonad: the fate of the supporting cell lineage. Bioessays. 1991, 13: 151-156.View ArticlePubMedGoogle Scholar
- Ottolenghi C, Uda M, Crisponi L, Omari S, Cao A, Forabosco A, Schlessinger D: Determination and stability of sex. Bioessays. 2007, 29: 15-25.View ArticlePubMedGoogle Scholar
- Ottolenghi C, Pelosi E, Tran J, Colombino M, Douglass E, Nedorezov T, Cao A, Forabosco A, Schlessinger D: Loss of Wnt4 and Foxl2 leads to female-to-male sex reversal extending to germ cells. Hum Mol Genet. 2007, 16: 2795-2804.View ArticlePubMedGoogle Scholar
- Polanco JC, Koopman P: Sry and the hesitant beginnings of male development. Dev Biol. 2007, 302: 13-24.View ArticlePubMedGoogle Scholar
- McElreavey K, Vilain E, Abbas N, Herskowitz I, Fellous M: A regulatory cascade hypothesis for mammalian sex determination: SRY represses a negative regulator of male development. Proc Natl Acad Sci USA. 1993, 90: 3368-3372.PubMed CentralView ArticlePubMedGoogle Scholar
- Eicher EM, Washburn LL: Genetic control of primary sex determination in mice. Annu Rev Genet. 1986, 20: 327-60.View ArticlePubMedGoogle Scholar
- Chassot AA, Ranc F, Gregoire EP, Roepers-Gajadien HL, Taketo MM, Camerino G, de Rooij DG, Schedl A, Chaboissier MC: Activation of beta-catenin signaling by Rspo1 controls differentiation of the mammalian ovary. Hum Mol Genet. 2008, 17: 1264-1277.View ArticlePubMedGoogle Scholar
- Kocer A, Pinheiro I, Pannetier M, Renault L, Parma P, Radi O, Kim KA, Camerino G, Pailhoux E: R-spondin1 and FOXL2 act into two distinct cellular types during goat ovarian differentiation. BMC Dev Biol. 2008, 8: 36-PubMed CentralView ArticlePubMedGoogle Scholar
- Smith CA, Shoemaker CM, Roeszler KN, Queen J, Crews DP, Sinclair AH: Cloning and expression of R-Spondin1 in different vertebrates suggests a conserved role in ovarian development. BMC Dev Biol. 2008, 8: 72-PubMed CentralView ArticlePubMedGoogle Scholar
- Maatouk DM, Dinapoli L, Alvers A, Parker KL, Taketo MM, Capel B: Stabilization of beta-catenin in XY gonads causes male-to-female sex-reversal. Hum Mol Genet. 2008, 17: 2949-2955.PubMed CentralView ArticlePubMedGoogle Scholar
- Crisponi L, Deiana M, Loi A, Chiappe F, Uda M, Amati P, Bisceglia L, Zelante L, Nagaraja R, Porcu S, Ristaldi MS, Marzella R, Rocchi M, Nicolino M, Lienhardt-Roussie A, Nivelon A, Verloes A, Schlessinger D, Gasparini P, Bonneau D, Cao A, Pilia G: The putative forkhead transcription factor FOXL2 is mutated in blepharophimosis/ptosis/epicanthus inversus syndrome. Nat Genet. 2001, 27: 159-166.View ArticlePubMedGoogle Scholar
- Uda M, Ottolenghi C, Crisponi L, Garcia JE, Deiana M, Kimber W, Forabosco A, Cao A, Schlessinger D, Pilia G: Foxl2 disruption causes mouse ovarian failure by pervasive blockage of follicle development. Hum Mol Genet. 2004, 13: 1171-1181.View ArticlePubMedGoogle Scholar
- Pailhoux E, Vigier B, Chaffaux S, Servel N, Taourit S, Furet JP, Fellous M, Grosclaude F, Cribiu EP, Cotinot C, Vaiman D: A 11.7-kb deletion triggers intersexuality and polledness in goats. Nat Genet. 2001, 29: 453-458.View ArticlePubMedGoogle Scholar
- Parma P, Radi O, Vidal V, Chaboissier MC, Dellambra E, Valentini S, Guerra L, Schedl A, Camerino G: R-spondin1 is essential in sex determination, skin differentiation and malignancy. Nat Genet. 2006, 38: 1304-1309.View ArticlePubMedGoogle Scholar
- Tibshirani R, Hastie T, Narasimhan B, Chu G: Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Natl Acad Sci USA. 2002, 99: 6567-6572.PubMed CentralView ArticlePubMedGoogle Scholar
- Cole SW, Galic Z, Zack JA: Controlling false-negative errors in microarray differential expression analysis: a PRIM approach. Bioinformatics. 2003, 19: 1808-1816.View ArticlePubMedGoogle Scholar
- Li H, Bishop KM, O'Leary DD: Potential target genes of EMX2 include Odz/Ten-M and other gene families with implications for cortical patterning. Mol Cell Neurosci. 2006, 33: 136-149.View ArticlePubMedGoogle Scholar
- Worzfeld T, Puschel AW, Offermanns S, Kuner R: Plexin-B family members demonstrate non-redundant expression patterns in the developing mouse nervous system: an anatomical basis for morphogenetic effects of Sema4D during development. Eur J Neurosci. 2004, 19: 2622-2632.View ArticlePubMedGoogle Scholar
- Byskov AG, Guoliang X, Andersen CY: The cortex-medulla oocyte growth pattern is organized during fetal life: an in-vitro study of the mouse ovary. Mol Hum Reprod. 1997, 3: 795-800.View ArticlePubMedGoogle Scholar
- Cvoro A, Tzagarakis-Foster C, Tatomer D, Paruthiyil S, Fox MS, Leitman DC: Distinct roles of unliganded and liganded estrogen receptors in transcriptional repression. Mol Cell. 2006, 21: 555-564.View ArticlePubMedGoogle Scholar
- Hara A, Inoue Y, Nakagawa M, Naganeo F, Sawada H: Purification and characterization of NADP+-dependent 3 alpha-hydroxysteroid dehydrogenase from mouse liver cytosol. J Biochem. 1988, 103: 1027-1034.PubMedGoogle Scholar
- Morita Y, Perez GI, Paris F, Miranda SR, Ehleiter D, Haimovitz-Friedman A, Fuks Z, Xie Z, Reed JC, Schuchman EH, Kolesnick RN, Tilly JL: Oocyte apoptosis is suppressed by disruption of the acid sphingomyelinase gene or by sphingosine-1-phosphate therapy. Nat Med. 2000, 6: 1109-1114.View ArticlePubMedGoogle Scholar
- Pan H, O'Brien MJ, Wigglesworth K, Eppig JJ, Schultz RM: Transcript profiling during mouse oocyte development and the effect of gonadotropin priming and development in vitro. Dev Biol. 2005, 286: 493-506.View ArticlePubMedGoogle Scholar
- Nef S, Schaad O, Stallings NR, Cederroth CR, Pitetti JL, Schaer G, Malki S, Dubois-Dauphin M, Boizet-Bonhoure B, Descombes P, Parker KL, Vassalli JD: Gene expression during sex determination reveals a robust female genetic program at the onset of ovarian development. Dev Biol. 2005, 287: 361-377.View ArticlePubMedGoogle Scholar
- Herrera L, Ottolenghi C, Garcia-Ortiz JE, Pellegrini M, Manini F, Ko MS, Nagaraja R, Forabosco A, Schlessinger D: Mouse ovary developmental RNA and protein markers from gene expression profiling. Dev Biol. 2005, 279: 271-290.View ArticlePubMedGoogle Scholar
- Pellegrini M, Pantano S, Lucchini F, Fumi M, Forabosco A: Emx2 developmental expression in the primordia of the reproductive and excretory systems. Anat Embryol (Berl). 1997, 196: 427-433.View ArticleGoogle Scholar
- Miyamoto N, Yoshida M, Kuratani S, Matsuo I, Aizawa S: Defects of urogenital development in mice lacking Emx2. Development. 1997, 124: 1653-1664.PubMedGoogle Scholar
- Choi Y, Ballow DJ, Xin Y, Rajkovic A: Lim Homeobox Gene, LHX8, Is Essential for Mouse Oocyte Differentiation and Survival. Biol Reprod. 2008, 79: 442-449.PubMed CentralView ArticlePubMedGoogle Scholar
- Coveney D, Ross AJ, Slone JD, Capel B: A microarray analysis of the XX Wnt4 mutant gonad targeted at the identification of genes involved in testis vascular differentiation. Gene Expr Patterns. 2007, 7: 82-92.View ArticlePubMedGoogle Scholar
- Gallardo TD, John GB, Shirley L, Contreras CM, Akbay EA, Haynie JM, Ward SE, Shidler MJ, Castrillon DH: Genomewide discovery and classification of candidate ovarian fertility genes in the mouse. Genetics. 2007, 177: 179-94.PubMed CentralView ArticlePubMedGoogle Scholar
- Joshi S, Davies H, Sims LP, Levy SE, Dean J: Ovarian gene expression in the absence of FIGLA, an oocyte-specific transcription factor. BMC Dev Biol. 2007, 7: 67-PubMed CentralView ArticlePubMedGoogle Scholar
- Soyal SM, Amleh A, Dean J: FIGalpha, a germ cell-specific transcription factor required for ovarian follicle formation. Development. 2000, 127: 4645-4654.PubMedGoogle Scholar
- Vainio S, Heikkila M, Kispert A, Chin N, McMahon AP: Female development in mammals is regulated by Wnt-4 signalling. Nature. 1999, 397: 405-409.View ArticlePubMedGoogle Scholar
- Govoroun MS, Pannetier M, Pailhoux E, Cocquet J, Brillard JP, Couty I, Batellier F, Cotinot C: Isolation of chicken homolog of the FOXL2 gene and comparison of its expression patterns with those of aromatase during ovarian development. Dev Dyn. 2004, 231: 859-870.View ArticlePubMedGoogle Scholar
- Wang DS, Kobayashi T, Zhou LY, Paul-Prasanth B, Ijiri S, Sakai F, Okubo K, Morohashi K, Nagahama Y: Foxl2 up-regulates aromatase gene transcription in a female-specific manner by binding to the promoter as well as interacting with ad4 binding protein/steroidogenic factor 1. Mol Endocrinol. 2007, 21: 712-725.View ArticlePubMedGoogle Scholar
- Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP: Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA. 2005, 102: 15545-1550.PubMed CentralView ArticlePubMedGoogle Scholar
- Hersmus R, Kalfa N, de Leeuw B, Stoop H, Oosterhuis JW, de Krijger R, Wolffenbuttel KP, Drop SL, Veitia RA, Fellous M, Jaubert F, Looijenga LH: FOXL2 and SOX9 as parameters of female and male gonadal differentiation in patients with various forms of disorders of sex development (DSD). J Pathol. 2008, 215: 31-38.View ArticlePubMedGoogle Scholar
- Batista F, Vaiman D, Dausset J, Fellous M, Veitia RA: Potential targets of FOXL2, a transcription factor involved in craniofacial and follicular development, identified by transcriptomics. Proc Natl Acad Sci USA. 2007, 104: 3330-3335.PubMed CentralView ArticlePubMedGoogle Scholar
- Guigon CJ, Magre S: Contribution of germ cells to the differentiation and maturation of the ovary: insights from models of germ cell depletion. Biol Reprod. 2006, 74: 450-458.View ArticlePubMedGoogle Scholar
- Jorgensen JS, Gao L: Irx3 is differentially up-regulated in female gonads during sex determination. Gene Expr Patterns. 2005, 5: 756-762.View ArticlePubMedGoogle Scholar
- Small CL, Shima JE, Uzumcu M, Skinner MK, Griswold MD: Profiling gene expression during the differentiation and development of the murine embryonic gonad. Biol Reprod. 2005, 72: 492-501.PubMed CentralView ArticlePubMedGoogle Scholar
- Beverdam A, Koopman P: Expression profiling of purified mouse gonadal somatic cells during the critical time window of sex determination reveals novel candidate genes for human sexual dysgenesis syndromes. Hum Mol Genet. 2006, 15: 417-431.View ArticlePubMedGoogle Scholar
- Bouma GJ, Affourtit JP, Bult CJ, Eicher EM: Transcriptional profile of mouse pre-granulosa and Sertoli cells isolated from early-differentiated fetal gonads. Gene Expr Patterns. 2007, 7: 113-123.View ArticlePubMedGoogle Scholar
- Armstrong SA, Staunton JE, Silverman LB, Pieters R, den Boer ML, Minden MD, Sallan SE, Lander ES, Golub TR, Korsmeyer SJ: MLL translocations specify a distinct gene expression profile that distinguishes a unique leukemia. Nat Genet. 2002, 30: 41-47.View ArticlePubMedGoogle Scholar
- Bair E, Tibshirani R: Semi-supervised methods to predict patient survival from gene expression data. PLoS Biol. 2004, 2: E108-PubMed CentralView ArticlePubMedGoogle Scholar
- Brunet JP, Tamayo P, Golub TR, Mesirov JP: Metagenes and molecular pattern discovery using matrix factorization. Proc Natl Acad Sci USA. 2004, 101: 4164-4169.PubMed CentralView ArticlePubMedGoogle Scholar
- Shi L, Reid LH, Jones WD, Shippy R, Warrington JA, Baker SC, Collins PJ, de Longueville F, Kawasaki ES, Lee KY, Luo Y, Sun YA, Willey JC, Setterquist RA, Fischer GM, Tong W, Dragan YP, Dix DJ, Frueh FW, Goodsaid FM, Herman D, Jensen RV, Johnson CD, Lobenhofer EK, Puri RK, Schrf U, Thierry-Mieg J, Wang C, Wilson M, Wolber PK, Zhang L, Amur S, Bao W, Barbacioru CC, Lucas AB, Bertholet V, Boysen C, Bromley B, Brown D, Brunner A, Canales R, Cao XM, Cebula TA, Chen JJ, Cheng J, Chu TM, Chudin E, Corson J, Corton JC, Croner LJ, Davies C, Davison TS, Delenstarr G, Deng X, Dorris D, Eklund AC, Fan XH, Fang H, Fulmer-Smentek S, Fuscoe JC, Gallagher K, Ge W, Guo L, Guo X, Hager J, Haje PK, Han J, Han T, Harbottle HC, Harris SC, Hatchwell E, Hauser CA, Hester S, Hong H, Hurban P, Jackson SA, Ji H, Knight CR, Kuo WP, LeClerc JE, Levy S, Li QZ, Liu C, Liu Y, Lombardi MJ, Ma Y, Magnuson SR, Maqsodi B, McDaniel T, Mei N, Myklebost O, Ning B, Novoradovskaya N, Orr MS, Osborn TW, Papallo A, Patterson TA, Perkins RG, Peters EH, Peterson R, Philips KL, Pine PS, Pusztai L, Qian F, Ren H, Rosen M, Rosenzweig BA, Samaha RR, Schena M, Schroth GP, Shchegrova S, Smith DD, Staedtler F, Su Z, Sun H, Szallasi Z, Tezak Z, Thierry-Mieg D, Thompson KL, Tikhonova I, Turpaz Y, Vallanat B, Van C, Walker SJ, Wang SJ, Wang Y, Wolfinger R, Wong A, Wu J, Xiao C, Xie Q, Xu J, Yang W, Zhong S, Zong Y, Slikker W: The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements. Nat Biotechnol. 2006, 24: 1151-1161.View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.